Quantification of Deep Traps in Nanocrystal Solids, Their Electronic

Oct 28, 2013 - Department of Information Technology and Electrical Engineering, ETH Zurich, Gloriastr. 35, 8092 Zurich ..... Charge carrier loss mecha...
0 downloads 10 Views 1MB Size
Letter pubs.acs.org/NanoLett

Quantification of Deep Traps in Nanocrystal Solids, Their Electronic Properties, and Their Influence on Device Behavior Deniz Bozyigit, Sebastian Volk, Olesya Yarema, and Vanessa Wood* Department of Information Technology and Electrical Engineering, ETH Zurich, Gloriastr. 35, 8092 Zurich, Switzerland S Supporting Information *

ABSTRACT: We implement three complementary techniques to quantify the number, energy, and electronic properties of trap states in nanocrystal (NC)based devices. We demonstrate that, for a given technique, the ability to observe traps depends on the Fermi level position, highlighting the importance of a multitechnique approach that probes trap coupling to both the conduction and the valence bands. We then apply our protocol for characterizing traps to quantitatively explain the measured performances of PbS NC-based solar cells. KEYWORDS: Deep level transient spectroscopy (DLTS), thermal admittance spectroscopy (TAS), Fourier transform photocurrent spectroscopy (FTPS), quantum dots, PbS, photovoltaics

S

emiconductor (SC) materials derived from colloidally synthesized nanocrystals (NCs) are of high interest for applications in third-generation photovoltaics (PV), due to their promise of reduced production costs and potential for high power conversion efficiencies.1−3 The current performance of NC solar cells still falls short of the requirement of 11− 14% power conversion efficiency target needed to justify commercialization.3 The low performance values for NC solar cells are often related to the presence of electronic trap states in the semiconducting NC solid, which act as recombination centers.4−6 Chemical treatments aiming to passivate such trap states have improved solar cell performance and have led to the demonstration of 7.4% efficiency using PbS NCs.6 While most reports agree that trap states are present and limit performance, only recently has there been an effort to understand and quantify trap states experimentally6−9 and theoretically.6,10,11 In particular, the impact of trap states on basic electrical measurements such as current−voltage (IV) and capacitance−voltage (or admittance spectroscopy) remains poorly understood. Because such measurements are the most heavily employed for characterization of NC-based devices, understanding how trap states manifest themselves in these measurements would enable the systematic improvement of the NC materials and thin films to achieve high performance devices. Here, we demonstrate and validate a protocol for rapid characterization of trap states in NC solids that provides quantitative information about the number and energy of the trap states and relate our findings to key solar cell performance metrics such as open circuit voltage and short circuit current. In our previous work, we applied current-based deep level transient spectroscopy (Q-DLTS) to a PbS NC solar cell (Figure 1a) and identified an abundant trap state manifold (T1) with an activation energy of ET1 = 0.40 eV.9 In this work, we develop a consistent picture (Figure 1b) of the bulk traps in NC solids and their influence on the electronic behavior of a © 2013 American Chemical Society

Figure 1. (a) Schematic of the Schottky junction solar cell. (b) Schematic of the density of states in the band gap of the PbS−EDT semiconductor, with the conduction band (blue), valence band (red), and the T1 trap manifold (dark orange). The experimental methods to determine energy offsets are indicated in gray. At day 0 the T1 traps are largely unoccupied, but a small change in the position of the Fermi level (EF) over two days in air increases the occupation of the T1 traps.

solar cell using Q-DLTS, thermal admittance spectroscopy (TAS), and Fourier transform photocurrent spectroscopy (FTPS) in conjunction with current voltage (IV) and photoluminescence (PL) measurements. For the experiments, we fabricate solar cells from PbS NCs in a diode structure with an ITO bottom anode and a LiF/Al/Ag top cathode (Figure 1a). The PbS NC thin film is deposited in air using a dipcoating procedure with an ethanedithiol (EDT) cross-linking step. The NCs used in this study show PL at 953 nm (1.30 eV) in solution (see Figure S1). Details of NC synthesis and the device fabrication procedure are given in ref 9. In the first step, we characterize the PV performance of our device by IV measurements under AM1.5G illumination. Our devices exhibit an open-circuit voltage of Voc = 0.50 ± 0.06 V, Received: July 27, 2013 Revised: October 18, 2013 Published: October 28, 2013 5284

dx.doi.org/10.1021/nl402803h | Nano Lett. 2013, 13, 5284−5288

Nano Letters

Letter

short-circuit current of Jsc = 8.67 ± 0.67 mA/cm2, fill factor of FF = 0.515 ± 0.006, and power conversion efficiency of ηPCE = 2.25 ± 0.18% with a fabrication yield of about 90%. The uncertainties are obtained from the measurement of seven equivalent devices (see Figure S2) and demonstrate that we are working with very reproducible devicesa prerequisite for the investigation of the device physics. To validate the quantitative agreement of the different trap characterization techniques, all data presented in this study is taken from one single device. First, we determine the trap state density using Q-DLTS.9 In such a measurement, the device is left at 0 V bias for 1 ms after which it is reverse-biased (−0.5 V) for 19 ms. The trapped charge is determined by integrating the current transient flowing out of the device. More details on the measurement technique are given in the Supporting Information and ref 9). Figure 2a shows the Q-DLTS spectra (ΔQ) for temperatures

later during which time the sample is stored in air (day 2 measurement), we find that the trapped charge increases 6-fold to 51 nC/cm2 (Figure 2b). This corresponds to a T1 trap density of 1.8 × 1017 cm−3 and an activation energy of ET1 = 0.36 eV. The discrepancy to the previously reported9 0.40 eV results from an improved data analysis method, rather than a change in the measurement results. The improved method is discussed in the Supporting Information and is validated by the excellent agreement to the activation energy measured using TAS as presented below. Next, we determine how the capacitance of the diode is influenced by the T1 trap. The capacitance is measured at 0 V bias with an impedance analyzer using a modulation amplitude of 10 mV (more details are given in the Supporting Information). We perform this measurement at temperatures between 160 K (blue) and 310 K (red). This is known as thermal admittance spectroscopy (TAS). The real part of the capacitance is plotted in Figure 2c. At high frequencies, we observe a capacitance of 175nF/cm2. We relate this value to the geometric capacitance Cgeo = ϵrϵ0/d, where ϵ0 is the electric constant, because, for d = 70 nm, we obtain a relative dielectric constant of ϵr = 14, which is in good agreement with other reports.12,13 On day 0, we observe a small increase in capacitance of up to 65 nF/cm2 at frequencies below 1 kHz. Figure 2d shows the same measurement on day 2, where the additional low frequency capacitance is much larger and exceeds 300 nF/cm2. This capacitance increase at low frequencies is indicative of a deep trap state. A method developed by Walter et al.14 allows the reconstruction of the density of trap states in the band gap (NT) from TAS measurements (see Figure 2e). In Figure S3 we plot the derivative the TAS data to determine the reduced attempt-frequency of ν00 = 1.2 × 105 s−1·K−2 (or an attemptfrequency of ν0 = 1.07 × 1010 s−1 at 300 K). The attemptfrequency ν0 represents how often a trapped carrier tries to escape to a conduction band state and is therefore an important metric for describing how a trap state interacts with the host semiconductor material. A previous attempt on TAS on PbS NC solids did not measure the attempt-frequency but estimated it to be on the order of 108 s−1.6 In other thin-film absorber materials, such as CIGS, attempt frequencies are often significantly larger (1011−1012 s−1).14 This implies that the PbS NC thin-film has a lower density of conduction band states, that the traps interact more weakly with the conduction band, or a combination of both. Using the TAS data and the attempt frequency, we plot in Figure 2e the trap density (NT) versus energy depth from the conduction band, assuming a built-in voltage of 0.6 V. The data shows a discrete trap state, which we fit by a Gaussian distribution (shading in Figure 2e). From the fit we obtain a density of NT1 = 3.0 × 1016 cm−3 and ET1 = 0.39 eV on day 0 and NT1 = 1.8 × 1017 and ET1 = 0.37 eV on day 2, in excellent agreement with the Q-DLTS values. The width of the Gaussian distributions are on the order of the thermal energy (σ = 20−28 meV), so we conclude that thermal broadening is limiting the energy resolution of the measurement. We note that the systematic uncertainty on the energy scale is on the order of 25 meV due to the uncertainties in the measured sample temperature. In addition to the discrete T1-trap manifold, NT features a plateau that extends further into the middle of the band gap (0.4−0.5 eV). These states also appear in the DLTS measurements as the broad shoulder at long time constants in high-temperature spectra (see Figure 2b). The fact that this

Figure 2. Q-DLTS spectrum for measurements on the device (a) immediately after fabrication and (b) after storage in air for two days. Spectra from 300 K (red) to 184 K (blue) are plotted. Black dots indicate the peak position that is used to determine the activation energy of the trap. (c, d) Thermal admittance spectroscopy (TAS) data for an applied bias voltage of 0 V. (e) Trap state density (NT) determined by TAS. Fitting a Gaussian (shading) to the discrete trap state reveals a T1-trap state density of 3.0 × 1016 cm−3 at day 0 and 1.8 × 1017 cm−3 at day 2, centered at an energy of ET1 = 0.37 eV. This is in excellent agreement with the properties of the T1 manifold determined by Q-DLTS (2.7 × 1016 cm−3 at day 0 and 1.8 × 1017 cm−3 at day 2, with ET1 = 0.36 eV).

between 300 K (red) and 184 K (blue), exhibiting a peak charge of 7.4 nC/cm2. We measure a PbS film thickness d = 70 nm by atomic force microscopy to relate this charge to a T1 trap density of 2.7 × 1016 cm−3. This measurement, which we will refer to as the day 0 measurement, is performed under vacuum, following air exposure for 5 min, 1 h after evaporation of the top electrode. Repeating the same experiment two days 5285

dx.doi.org/10.1021/nl402803h | Nano Lett. 2013, 13, 5284−5288

Nano Letters

Letter

shoulder appears only at high temperatures at long time constants and is broad in shape is compatible with a distribution of midgap states. Scanning tunneling spectroscopy measurements on EDT-treated PbS NC-films have recently found a similar density of states.15 Further agreement between Q-DLTS and TAS is found in the value for the trap capture cross section (σT1). Like the attempt frequency, σT1 quantifies the interaction of the trap states with mobile carriers. It can be determined for both TAS and Q-DLTS measurements as shown in the Supporting Information. The cross sections derived from TAS (σT1,TAS = 3.7 × 10−16 cm2) and from DLTS (σT1,DLTS = 6.0 × 10−16 cm2) correspond to a disc with a capture radius of 110−140 pm. We note that this value is in the range of the ionic radius of most metal ions (50−130 pm),16 which suggests a possible origin for the observed traps. The fact that the extracted capture radius is physically reasonable justifies the use of the effective mass approximation in the Q-DLTS and TAS data analysis (see Supporting Information). Although this approximation was originally developed in the context of band transport semiconductors, our results show that it is applicable to NC solids to describe trap dynamics. On first consideration, the strong quantitative agreement of Q-DLTS and TAS measurements may not be surprising as both techniques are based on charge measurements in the dark. In fact, in some cases, time-domain and frequency-domain measurements are equivalent and are trivially related by the Fourier transform.17 However, for Q-DLTS and TAS, this is not the case. Q-DLTS is a large-signal technique where the total amount of trapped charge is directly measured (in units of cm−3). In contrast, TAS is a small-signal technique and requires an analytical model of the band bending in the device to provide the spectral density of trapped charge (in units of cm−3· eV−1). Therefore, the agreement in the quantity and energy of traps found with Q-DLTS and TAS is significant. Q-DLTS validates the assumptions needed for TAS analysis, which is an important step toward a consistent model of charge transport in NC-based devices. Traps are particularly detrimental to photovoltaic performance if they are coupled to both the valence band (VB) and the conduction band (CB) and therefore act as recombination centers. Q-DLTS and TAS measurements consistently show trap states with an activation energy of ET1 = 0.36 eV, which we interpret as the energy distance to the CB. Our measurements do not allow us to discriminate whether the T1 trap is close to the VB or the CB. We tend to think that it is close to the CB based on the measurements in ref 18. In contrast, ref 15 indicated that the dominant trap states are closer to the VB. We test the coupling of this T1 trap manifold to the VB using the Fourier transform photocurrent spectroscopy (FTPS) technique (see Figure 3a).19 In this technique, we measure the photocurrent spectrum by illuminating the device with infrared light from an FTIR spectrometer (more experimental details are given in the Supporting Information). The photocurrent plotted in Figure 3b shows a distinct peak at 0.67 eV. Given the band gap for this NC-solid of 1.0−1.2 eV determined by PL (Figure 3b red), 0.67 eV matches the expected energy for the T1−VB transition. We therefore conclude that the T1 traps couple efficiently to both the VB and the CB and can therefore act as recombination centers. The peak in the subbandgap photocurrent that we measure with FTPS is on the order of 5.3pA on day 0 (solid black line in Figure 3b) and reduces by 16% on day 2 (dashed black line).

Figure 3. (a) Experimental setup for FTPS measurements. Under infrared illumination, the photocurrent is measured and recorded by a Fourier transform infrared spectrometer (FTIR). (b) Photoluminescence spectrum of the NC-solid (red, right axis) and photocurrent (left axis) for day 0 (solid black line) and day 2 (dashed black line). PL peak at 1.15 eV determines the effective band gap of the semiconductor. The photocurrent shows a transition at 0.67 eV, which is consistent with the energetic position of the T1 trap measured with Q-DLTS and TAS.

This reduction is accompanied by a shift to higher energies that is on the order of 10 meV (see Figure S4b). The decrease in photocurrent magnitude and the shift in energy is surprising and at first seems to contradict the apparent increase in traps under air exposure observed with Q-DLTS and TAS measurements. However, a consistent picture for device behavior can be developed that uses the concept of a partially filled trap band proposed in ref 18. As sketched in Figure 1b, if, at day 0, the Fermi level is lower in energy than the center of the T1 trap manifold, the T1 trap manifold will be largely unoccupied. Indeed, through electrical measurements (Q-DLTS, TAS), we observe a relatively small density of traps at the 0.36 eV transition to the CB. A shift in Fermi level such that a greater fraction of traps is occupied in thermal equilibrium on day 2 than on day 0 explains the larger Q-DLTS and TAS signals measured on day 2. Furthermore, because this shift in Fermi level would also mean that fewer trap states are available for absorption to take place, the shift is consistent with the decrease and shift to higher energy of the FTPS signal. From our measurements of the trap occupation on day 0 (Nd0) and day 2 (Nd2), we can estimate a 50 meV change in the Fermi level (ΔEF) using Boltzmann statistics: Nd2/Nd0 = 6.7 = exp(ΔEF/kT), where kT is the thermal energy. This shift could arise from exposure to water or oxygen as the sample sits in air. We note that oxidation of the PbS would tend to lower the Fermi level instead of increase it.20 This suggests either (1) that the trap state manifold is closer to the VB and that Q-DLTS and TAS probe the VB to trap transition while FTPS probes the trap state to CB transition such that our results should be explain by a decrease in the Fermi level, or (2) that the change in Fermi energy is not due to oxidation of PbS but results from a more complex process. It was indicated by the work of Tang et al. that the degradation under air exposure is related to the Al electrode.21 Recently it has been shown that n-type PbS NC solids can be obtained by the contact with a reducing agent,22 which makes it plausible, that Al, as a strong reducing agent, could have a similar effect and slightly increase the Fermi energy upon degradation of the electrode. 5286

dx.doi.org/10.1021/nl402803h | Nano Lett. 2013, 13, 5284−5288

Nano Letters

Letter

VB and the CB and can act as recombination centers in our solar cell. We leverage the change in measured trap densities in our device as a function of time to show that device aging in air results in a shift in the Fermi level and trap occupation rather than in the creation of additional traps. This new perspective on device degradation is consistent with the change in the IV characteristics of the diode. We also show that the T1 trap states are responsible for a large increase (up to 300 nF/cm2) in the low frequency capacitance of the diode. This proves that the popular Mott−Schottky characterization technique is not in general applicable for NC solids.12,13,23 Determination of the space charge region requires a more advanced technique, such as drive-level capacitance profiling (DLCP), that takes the lowfrequency capacitance into account.24 Of the three measurement techniques we perform, TAS emerges as the most simple and robust technique to determine the density of states in the band gap of a solution processed NC solid. At the same time we find that trap states can be invisible to a TAS measurement depending on the occupation of the trap states due to the position of the Fermi level in the device. This can explain why earlier reports were not able to observe trap states using TAS.6 Only in combination with a technique that probes the complementary transitionsuch as FTPS can a definite account on the presence and quantity of trap states be given.

To understand how trap states influence the device performance, we perform current−voltage (IV) measurements. Figure 4a shows nearly identical IV characteristics on day 0



Figure 4. (a) Current−voltage characteristic in the dark at day 0 (solid) and at day 2 (dashed). (b) Current−voltage characteristic under AM1.5G illumination at day 0 (solid) and day 2 (dashed). We observe a slight reduction of the photocurrent at day 2, consistent with a change in the Fermi level in the semiconductor.

ASSOCIATED CONTENT

S Supporting Information *

Information on reproducibility of device fabrication, improved Q-DLTS data analysis, calculation of Q-DLTS and TAS parameters, additional information on FTPS measurements, and estimation of field reduction by trapped charges. This material is available free of charge via the Internet at http:// pubs.acs.org.

(solid) and day 2 (dashed). Under AM1.5G illumination (Figure 4b), we observe that the open-circuit voltage (Voc) does not change within the observation time of two days and that the short-circuit current (Jsc) decreases by 23% from 9.3 mA/cm2 to 7.2 mA/cm2. These observations are in line with the picture in which the occupation of the T1 trap manifold changes due to a shift in the Fermi level following exposure of the sample to air. First, the Voc is limited by internal recombination and sensitive to the number of trap states.3 We do not observe a change in Voc from day 0 to day 2, which supports our conclusion that the total number of traps does not change significantly. Second, the reduction in Jsc indicates a reduced built-in field (Fbi). When the occupation of the trap states in the semiconductor changes, it is thermodynamically required to reduce the built-in field and thereby reduce Jsc. Assuming that Jsc ∝ Fbi, we find that the observed reduction in Jsc by 23% would need a charge rearrangement of 24 nC/cm2 (see Supporting Information). This is comparable to the change in trapped charge (43.6 nC/ cm2), which we observe in the Q-DLTS measurements. Thus, the change in our current−voltage characteristics is consistent with the internal rearrangement of charge that is observed through the three distinct measurements of the occupied trap states. In summary, we have investigated trap states in PbS NCbased thin films and determined their influence on the electronic behavior of Schottky-type diodes. We implement three distinct measurementsQ-DLTS, TAS, and FTPSand demonstrate that all agree quantitatively, showing a deep trap manifold (T1) with a density of NT1 ≥ 1.8 × 1017 cm−3 and an activation energy to the CB of ET1 − ECB = 0.36 eV. To our knowledge, the FTPS measurements are the first on NC solids and demonstrate that the traps interact efficiently with both the



AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected], phone: +41-44-632 66 54. Author Contributions

D.B. fabricated devices and performed all measurements. S.V. built the FTPS measurement setup. O.Y. synthesized the materials. D.B. and V.W. envisioned the experiments. The manuscript was written through contributions of all authors. Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS This research was supported by the Swiss National Science Foundation through grant 200021_135432 and the National Centre of Competence in Research Quantum Science and Technology.



ABBREVIATIONS Q-DLTS, charge DLTS; TAS, thermal admittance spectroscopy; IV, current−voltage characteristics; FTPS, Fourier transform photocurrent spectroscopy; PL, photoluminescence; NC, nanocrystal; PV, photovoltaic; SC, semiconductor; EDT, ethanedithiol; ITO, indium−tin−oxide; VB, valence band; CB, conduction band; CIGS, copper indium gallium sulfide; DLCP, drive-level capacitance profiling 5287

dx.doi.org/10.1021/nl402803h | Nano Lett. 2013, 13, 5284−5288

Nano Letters



Letter

REFERENCES

(1) Sargent, E. H. Adv. Mater. 2008, 20, 3958−3964. (2) Talapin, D. V.; Lee, J.-S.; Kovalenko, M. V.; Shevchenko, E. V Chem. Rev. 2010, 110, 389−458. (3) Lunt, R. R.; Osedach, T. P.; Brown, P. R.; Rowehl, J. A.; Bulović, V. Adv. Mater. 2011, 23, 5712−27. (4) Kramer, I. J.; Sargent, E. H. ACS Nano 2011, 5, 8506−14. (5) Szendrei, K.; Speirs, M.; Gomulya, W.; Jarzab, D.; Manca, M.; Mikhnenko, O. V.; Yarema, M.; Kooi, B. J.; Heiss, W.; Loi, M. A. Adv. Funct. Mater. 2012, 22, 1598−1605. (6) Ip, A. H.; Thon, S. M.; Hoogland, S.; Voznyy, O.; Zhitomirsky, D.; Debnath, R.; Levina, L.; Rollny, L. R.; Carey, G. H.; Fischer, A.; Kemp, K. W.; Kramer, I. J.; Ning, Z.; Labelle, A. J.; Chou, K. W.; Amassian, A.; Sargent, E. H. Nat. Nanotechnol 2012, 7, 577−82. (7) Konstantatos, G.; Sargent, E. H. Appl. Phys. Lett. 2007, 91, 173505. (8) Erslev, P. T.; Chen, H.-Y.; Gao, J.; Beard, M. C.; Frank, A. J.; van de Lagemaat, J.; Johnson, J. C.; Luther, J. M. Phys. Rev. B 2012, 86, 155313. (9) Bozyigit, D.; Jakob, M.; Yarema, O.; Wood, V. ACS Appl. Mater. Interfaces 2013, 5, 2915−9. (10) Voznyy, O.; Thon, S. M.; Ip, A. H.; Sargent, E. H. J. Phys. Chem. Lett. 2013, 987−992. (11) Kim, D. D.-H.; Lee, J.-H.; Grossman, J. C. Phys. Rev. Lett. 2013, 110, 196802. (12) Luther, J. M.; Law, M.; Beard, M. C.; Song, Q.; Reese, M. O.; Ellingson, R. J.; Nozik, A. J. Nano Lett. 2008, 8, 3488−92. (13) Clifford, J. P.; Johnston, K. W.; Levina, L.; Sargent, E. H. Appl. Phys. Lett. 2007, 91, 253117. (14) Walter, T.; Herberholz, R.; Muller, C.; Schock, H. W. J. Appl. Phys. 1996, 80, 4411. (15) Diaconescu, B.; Padilha, L. A.; Nagpal, P.; Swartzentruber, B. S.; Klimov, V. I. Phys. Rev. Lett. 2013, 110, 127406. (16) Shannon, R. D. Acta Crystallogr., Sect. A 1976, 32, 751−767. (17) Barnes, P. R. F.; Miettunen, K.; Li, X.; Anderson, A. Y.; Bessho, T.; Gratzel, M.; O’Regan, B. C. Adv. Mater. 2013, 25, 1881−1922. (18) Nagpal, P.; Klimov, V. I. Nat. Commun. 2011, 2, 486. (19) Vanecek, M.; Poruba, A. Appl. Phys. Lett. 2002, 80, 719. (20) Zarghami, M. H.; Liu, Y.; Gibbs, M.; Gebremichael, E.; Webster, C.; Law, M. ACS Nano 2010, 4, 2475−85. (21) Tang, J.; Wang, X.; Brzozowski, L.; Barkhouse, D. A. R.; Debnath, R.; Levina, L.; Sargent, E. H. Adv. Mater. 2010, 22, 1398− 402. (22) Koh, W.; Koposov, A. Y.; Stewart, J. T.; Pal, B. N.; Robel, I.; Pietryga, J. M.; Klimov, V. I. Sci. Rep. 2013, 3, 2004. (23) Brown, P. R.; Lunt, R. R.; Zhao, N.; Osedach, T. P.; Wanger, D. D.; Chang, L.-Y.; Bawendi, M. G.; Bulović, V. Nano Lett. 2011, 11, 2955−61. (24) Michelson, C. E.; Gelatos, A. V.; Cohen, J. D. Appl. Phys. Lett. 1985, 47, 412.

5288

dx.doi.org/10.1021/nl402803h | Nano Lett. 2013, 13, 5284−5288